2. The data

The database has been constructed in the first place from official national statistics giving annual numbers of deaths by sex and single years of age, beginning with the age of 80 years. In most countries, the single-year classification by age continues till the highest age recorded while in several others, the deaths at ages such as 100 and over are given in a lump sum which naturally impairs their usefulness. Often, however, this is done because the detailed age data are considered unreliable by the national office.

In about half of the countries of the database the age at death is cross-tabulated with the year of birth (cohort) and this information is used for arranging the data by cohort in order to make use of the extinct cohorts method. A problem rises when the cohort is not given and the deaths at a given age have to be allocated to the two alternative cohorts by an arbitrary rule. This was done at first in proportions observed in a neighbouring country but has later been done at a 50/50 split because the choice makes no appreciable difference in decennial life tables for either periods or cohorts. On the other hand, no arbitrary cohort allocation can lead to satisfactory results when calculating death rates by this method for one-year periods or for single cohorts. For this, double classification by age/cohort is needed and it is available for the following 15 countries: Belgium, Czechoslovakia, Denmark, Finland, France, West Germany, Hungary, Iceland, Italy, Japan, the Netherlands, Poland, Spain, Sweden and Switzerland. In Norway, the data were classified until 1976 by cohort only, thereafter by age only.

As mentioned, the data on deaths are in several countries limited by the fact that the single-year classification ends at age 99 or 100 after which the older deceased are given in a lump sum. This is the case for Canada, Czechoslovakia, Estonia, East Germany, Hungary, Latvia, Luxembourg, Poland and Scotland for the entire period, for West Germany until 1963, Australia until 1967, Belgium until 1973 and Singapore until 1988. The data for Spain are similarly limited since 1981 while Estonia removed the age restriction in 1990.

The official death counts have been accepted in the database virtually without correction. The deaths at ages 110 and over in England and Wales include only cases verified against a birth registration.

When a cohort is judged extinct, the numbers of deaths are summed up beginning with the oldest, to give the number of persons who reached each successive age. This ignores international migration which has been found to be negligible at high ages. The same procedure is applied even for countries which produce accurate migration data in their population registers, the reason being the methodological advantage of interlocking data on population and deaths.

For cohorts which are not yet extinct, the number or current survivors is needed in order to obtain the populations-at-risk. To determine these numbers, official population data have been compared with our own estimates made under the assumption that recent age-specific survival ratios in two or more successive cohorts have been equal or have changed linearly. Official population data are in these countries generally very reliable except at ages approaching or exceeding 100 years when they often become unreliable and even implausibly high. In such cases, the "own estimate" is accepted because, though not exact, it is consistent with information on deaths and generally plausible. With younger ages the "own estimates" become increasingly uncertain while the official figures gain in accuracy. Therefore, as a rule, a switch is made to the official figures at some such age as 95 or 90, often made easy by a convergence of the two series. However, for countries with reliable population registers, the official data are accepted as such.

This method of estimating survivors has been tested by applying it to past situations which can be verified by post-facto evidence and it has been found quite satisfactory.

On the basis of rigorous quality tests, the data were classified into the following four categories:

*	Austria	*	Iceland
	Belgium	*	Italy
	Czechoslovakia	*	Japan
*	Denmark		Luxembourg
*	England & Wales	*	Netherlands
*	Finland	*	Norway
*	France		Scotland
	Germany, East	*	Sweden
*	Germany, West	*	Switzerland
	Hungary

Thirteen of the nineteen countries above, marked with an asterisk, have been combined into a hard core of best information because in them the age at death is known without an upper limit. This group of thirteen, though not uniform, shares a low level of mortality as well as the experience of a substantial recent decline in it. It will therefore be frequently used for measurement of various aspects of old age survival and their recent developments.

In the other six countries, the detailed classification ends at 99 or 100; in Belgium this was the case until 1973.

Australia

New Zealand, non-Maori

Portugal

Singapore, Chinese

The data for Australia and New Zealand carry traces of possible age overstatement which may now have come under control. All data presented in this study for New Zealand exclude the Maori. The Portuguese data have improved over time and have been in the last two decades quite reliable except at ages above 105. The series for Singapore is as yet too short for conclusive evaluation. Data for other races in Singapore are available but the numbers are too small for analysis.

These data give probably a roughly correct description of the mortality trend though at a level artificially lowered by age overstatement.

Canada

Chile

New Zealand, Maori

United States

Used with caution, these data may give approximate information on the size and development of the population below age 90 but estimation of mortality by the extinct cohort method would be too uncertain. The data for New Zealand Maori and United States non-white are the weakest due to large-scale age overstatement.

The "World Population Data Sheet" of the Population Reference Bureau Inc. (1994) lists 35 countries or territories in which the life expectancy at birth for both sexes combined is 75 years or more. These have a total population of 838 million which may be divided as follows:¹

core thirteen	434 million
five others	71 "
covered in this study	505 million

United States	261 "
Canada	29 "
others	43 "
total	838 million

If, as we believe, old age mortality in the United States and Canada is comparable to that in the "thirteen" of the database, these latter represent well the low-mortality countries of the world. However, if - as has been claimed but not proven - old age mortality is lower in the U.S. than anywhere else, then the representativeness is of course greatly limited. Whichever case one chooses to believe, the database represents the cutting edge of the decline in old age mortality as far as it can be accurately measured. This holds true particularly of the core of thirteen, composed as it is of exactly the countries with the lowest proven levels of mortality among the oldest-old.

The combined population of the 28 countries included in the present study is according to the same source 590 million and the life expectancy in all of them is at least 70 years with the exception of Hungary and Latvia where it is 69.

The European countries of the database are further characterized by an advanced degree of aging. The fourteen countries in which according to the U.S. Bureau of the Census (1991) the proportion aged 75 and over is the highest in the world, includes eleven of the thirteen (without Iceland and Japan) plus Belgium, Greece and Luxembourg.

The size of the population aged 80 and over in 1950-1990 is given by sex, age and country in Annex Table 1. The total volume of the data used in the present study amounted to 386 million person-years of observation of which 52 million ended in death as summarized in Table 1. The bulk of the information, 333 million person-years-at-risk (86.9 percent) refers to the 19 countries with good quality data, and 287 million (74.9 percent) to the core of thirteen countries with unrestricted age information in the 1960-90 period.

The rapid attrition of the oldest-old population by death leads to a very uneven age distribution. Persons aged 90-94 are little more than one-tenth of those ten years younger but are almost 100 times as numerous as the centenarians. The analytically and scientifically important centenarian group is therefore only a tiny fraction of the total - yet, the nearly 300,000 life years which they began and of which nearly 122,000 ended in death, constitute probably a larger body of reliable evidence of survival after age 100 than has been assembled before.

In population terms, almost exactly one-third of the evidence pertains to the male but due to their higher mortality, they account for more than 38 percent of the deaths.

In the short historical period encompassed by the study, the oldest-old population grew vigorously. A part of the increase in Table 1 is due to inclusion of more countries in the course of the years. However, the greatest factor by far has been an increase in the number of persons surviving to high ages in the countries studied.

The mean annual probability of dying, though only a rough summary measure, reflects the steep increase of mortality by age, the greater vulnerability of men, and an unmistakable decline over time.

1.	United Kingdom and Germany are each counted as one entity; data for the U.S., Canada and Chile are entered in the database but were only exceptionally made use of in the study.